华中师范大学学报(自然科学版)2026,Vol.60Issue(5):793-804,12.DOI:10.19603/j.cnki.1000-1190.2026.05.008
融合多层次分割与XGBoost的青海省地貌分类研究
Landform classification of Qinghai Province using integrated multi-level segmentation and XGBoost
摘要
Abstract
Traditional single-scale object-based land form classification methods are prone to misclassification and omission,and the resulting segmentation units often lack clear geomorphological meaning.Taking Qinghai Province as the study area,this study proposes a land form classification framework for complex plateau land form regions by integrating multi-level slope unit segmentation,feature selection,and extreme gradient boosting(XGBoost).The framework uses a multi-level segmentation strategy to extract slope units,calculates five categories of features,including morphology,structure,physical properties,texture,and shape,and applies the shapley additive explanations(SHAP)method to identify key features for land form classification.The results show that the framework achieves an overall accuracy of 92%and an average F1 score of 0.91,which are 4 percentage points and 0.03 higher,respectively,than those of the single-scale method,and it outperforms Light Gradient Boosting Machine(LightGBM)and Random Forest models.The framework achieves accurate identification of land form types such as high-altitude plains,mid-altitude plains,high-altitude hills,and moderately high undulating mountains.It can effectively reflect the spatial heterogeneity and reasonable boundaries of land forms,and provides an effective methodological framework for automated land form mapping in large-scale complex plateau terrain regions.关键词
地貌划分/斜坡单元/多尺度分割/XGBoostKey words
land form classification/slope unit/multi-scale segmentation/XGBoost分类
天文与地球科学引用本文复制引用
王煜,江文萍,王玥,戴丹凤,范佳薇,徐梦琪..融合多层次分割与XGBoost的青海省地貌分类研究[J].华中师范大学学报(自然科学版),2026,60(5):793-804,12.基金项目
国家重点研发计划项目(2022YFB3904100) (2022YFB3904100)
国家自然科学基金项目(41371428) (41371428)
武汉大学测绘遥感信息工程国家重点实验室专项科研经费项目. ()